AnonymousSite/QA_dataset_for_CCLR

1. Introduction

(1) As far as we know, this is the largest QA dataset for Chinese Construction Laws and Regulations (CCLR). For example, well-known datasets like c-eval typically contain only about 500 questions in a single domain, whereas our dataset specifically focuses on the CCLR domain and includes 6,179 questions.

(2) This dataset has 2,060 questions from Registered Constructor Qualification Examination (RCQE) and 4,119 self-designed questions covering 8 CCLR subdomains.

(3) This dataset is developed and maintained by Southeast University, University of Cambridge, and City University of Hong Kong.

(4) Make sure to read the specification and follow the rules.

2. Submission of your LLM’s answers

The answers could be submitted through https://forms.gle/bKLj6GgyxSnGenXS8. Please use “Template of answer submission.xls” in this repository to submit your LLM's answers

3. Citation requirement

The reuse of this repository requires citation. Should any individual or entity utilize this repository without appropriate acknowledgment and citation, they do not have the right to use our data. We will take measures to protect our copyright, including, but not limited to, retracting their papers and initiating legal action.

4.LLM Leaderboard for CCLR QA

Large Language Model	Contributors	Overall Scoring Rate	D1	D2	D3	D4	D5	D6	D7	D8	Ranking
ERNIE-Bot 4.0 with knowledge graph	Baidu & The authors	0.822	0.850	0.828	0.836	0.803	0.843	0.844	0.800	0.860	1
ERNIE-Bot 4.0	Baidu	0.757	0.783	0.716	0.763	0.769	0.718	0.725	0.732	0.785	2
GPT-4 with knowledge graph	OpenAI & The authors	0.663	0.720	0.731	0.668	0.656	0.754	0.685	0.668	0.688	3
GPT-4	OpenAI	0.537	0.602	0.487	0.559	0.537	0.565	0.517	0.513	0.566	4
GPT-3.5-turbo with knowledge graph	OpenAI & The authors	0.503	0.532	0.505	0.523	0.468	0.613	0.522	0.544	0.464	5
ChatGLM3-6B with knowledge graph	Tsinghua, Zhipu.AI & The authors	0.483	0.497	0.450	0.509	0.428	0.536	0.499	0.545	0.445	6
Text-davinci-003 with knowledge graph	OpenAI & The authors	0.481	0.507	0.522	0.470	0.479	0.576	0.514	0.515	0.514	7
Qianfan-Chinese-Llama-2-7B with knowledge graph	Baidu & The authors	0.474	0.474	0.490	0.493	0.467	0.560	0.530	0.517	0.474	8
ChatGLM2-6B with knowledge graph	Tsinghua, Zhipu.AI & The authors	0.467	0.469	0.468	0.488	0.462	0.515	0.504	0.530	0.465	9
ChatGLM2-6B	Tsinghua & Zhipu.AI	0.427	0.452	0.411	0.475	0.412	0.461	0.447	0.492	0.421	10
ChatGLM3-6B	Tsinghua & Zhipu.AI	0.399	0.454	0.391	0.412	0.362	0.410	0.388	0.416	0.400	11
Qianfan-Chinese-Llama-2-7B	Baidu	0.373	0.419	0.380	0.368	0.360	0.415	0.376	0.416	0.357	12
GPT-3.5-turbo	OpenAI	0.346	0.425	0.316	0.362	0.326	0.432	0.332	0.406	0.334	13
Llama-2-70b with knowledge graph	MetaAI & The authors	0.377	0.336	0.368	0.320	0.331	0.411	0.354	0.337	0.336	14
Text-davinci-003	OpenAI	0.327	0.352	0.316	0.341	0.337	0.381	0.344	0.363	0.343	15
Llama-2-70b	MetaAI	0.284	0.285	0.338	0.255	0.316	0.312	0.288	0.302	0.295	16